Weights & Biases founder and CEO Lukas Biewald demos the company’s machine learning tools for Robot Brains Podcast host Pieter Abbeel.
Biewald walks through a Colab example: install and import the Python package, initialize a project with `wandb.init`, then log metrics such as accuracy and loss with `wandb.log`. The resulting dashboard compares runs, displays logged details such as GPU type and notes, and lets users customize graphs.
Biewald shows a KMNIST project with 512 runs, which can be sorted by hyperparameters or results such as loss. GPU utilization panels can reveal that a run is not using its GPU fully. Other visualizations explore relationships across hyperparameters; in his example, the runs with the highest validation accuracy had lower dropout.
Biewald describes tools for exploring logged data and model behavior. In an iNaturalist classification example, users can group predictions and inspect images where the model confused fungi with amphibians, as well as view score histograms. He also shows a parameter importance panel that uses a boosted tree model to estimate which hyperparameters matter for a chosen output.
For production workflows, Biewald says teams can trace how training inputs relate to upstream runs and evaluations, then inspect the logged details of each run. A model registry tracks what is entering or leaving production and supports model cards. Reports can bring charts and selections together with written analysis, inline equations, and LaTeX export, while linking conclusions back to runs, code, and loss curves.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
Geoffrey Hinton, AI researcher and former Google employee, talks with host Pieter Abbeel on The Robot Brains Podcast about AI’s growing capabilities, risks, potential benefits, and his decision to leave Google.
Nando de Freitas, research director at DeepMind, talks with Pieter Abbeel on The Robot Brains Podcast about making AI useful and accessible worldwide, generalist agents, robotics, and future research.
Pieter Abbeel, director of Berkeley’s Robot Learning Lab and co-director of BAIR, presents on deep learning for robotics at a CMU Robotics Institute seminar hosted by Dave.
